Usage instructions
Core usage
Capability Evolver is a meta-skill that gives AI Agent the ability to evolve itself. Users only need to runnode index.jsA fully automatic evolution cycle can be started. The system will automatically scan the runtime history, identify errors or inefficient patterns, and generate repair patches or optimized code. Supports three operating modes: standard fully automatic mode (Mad Dog Mode), manual review mode (--review) and background loop daemon mode (--loop）。
Significant advantages
1. Zero dependency design: Only Node.js built-in modules are used, with no third-party runtime dependencies, greatly reducing the risk of supply chain attacks.
2. Protocol constraint evolution: Adopting the GEP (Genome Evolution Protocol) protocol, all evolutionary events are traceable and auditable, and the integrity of assets is ensured through SHA-256 content addressing.
3. Multiple layers of security: Built-in command whitelist (only node/npm/npx allowed), path traversal protection, automatic desensitization of sensitive information, explosion radius control (up to 60 files/20,000 lines in a single cycle), and automatic rollback mechanism on failure.
4. Intelligent resource management: Singleton Guard prevents multi-instance conflicts, memory leak protection (RSS limit + automatic restart), saturation detection to avoid idling and wasting resources.
5. context-agnostic architecture: Implement local preference injection through environment variables or dynamic detection, and can adapt to different workflows without modifying the core code.
Potential Disadvantages and Limitations
1. The inherent uncertainty of self-modification: Despite the constraints of the GEP protocol, there is still the risk of unpredictable behavior when the Agent autonomously modifies its own code, and may produce side effects that are difficult to debug in complex scenarios.
2. The review threshold is higher：events.jsonlThe evolution event log requires professional knowledge to effectively audit, making it difficult for ordinary users to judge the quality of evolution.
3. Innovative cooling mechanisms may inhibit necessary refactoring: The cooling-off period set up to prevent repeated innovation of the same skill may become an obstacle in emergency architecture adjustment scenarios.
4. A2A asset dissemination relies on external verification: Required for external genetic enhancement--validatedSign, the trust establishment process when collaborating across Agents is more cumbersome.
Suitable target group
- AI Agent Developer: Need to build a long-running Agent system with self-maintenance capabilities
- Automated operation and maintenance team: Hope to reduce manual intervention in fault repair and performance tuning work
- research users: Exploring the safety boundaries and feasible solutions of recursive self-improvement systems
- OpenClaw Ecosystem User: Already using the OpenClaw framework and hoping to expand the Agent’s autonomous capabilities
Risks of use
1. Version control dependencies: It is strongly recommended to cooperate with git-sync for regular backups, otherwise the rollback when automatic repair fails may lose the intermediate state.
2. Sensitive environment configuration: Must be enabled in production environment--reviewmode to avoid unexpected changes caused by the real-time execution of Mad Dog Mode.
3. Memory and storage growth：memory//directory andevents.jsonlWill continue to grow and requires monitoring and regular archiving.
4. Cycle model operation and maintenance costs：--loopBackground operation requires cooperation with cron or process management tools. Improper configuration may lead to zombie processes or resource contention.
Safety review
Core usage
Capability EvolverIt is a meta-level skill that allows AI Agents to have the ability to self-examine and iterate. The core entrance isnode index.js, providing three operating modes:
- Fully automatic mode(Default/Mad Dog Mode): Directly execute the complete cycle of detection → analysis → repair → optimization
- censorship mode（
--review): Pause and wait for manual confirmation before each operation, suitable for production environment - Continuous cycle（
--loopor cron scheduling): run as a background daemon
Its bottom layer adoptsGEP Protocol (Genome Evolution Protocol), realizing an auditable and traceable evolution process through structured asset storage (genes.json gene definition, capsules.json success capsule, events.jsonl event log).
Significant advantages
1. Zero supply chain risk: No npm external dependencies, only using Node.js built-in modules, completely eliminating dependency poisoning attacks
2. Multi-layered security controls: Command whitelist (only node/npm/npx is allowed), dangerous mode detection, risk downgrade strategy (high-risk personality prohibits innovative operations), Git lock protection
3. Environment decoupling design:passEVOLVE_REPORT_TOOLInject local preferences into environment variables and keep the core code pure
4. Identity-driven intelligence: Built-in "recursive self-improvement system" identity injection, automatically switches to repair mode or forced optimization mode according to running status
Potential Disadvantages and Limitations
- Single process architecture limitations: Infinite recursion is explicitly prohibited, but this also means that the ability to explore complex problems in parallel is limited.
- Git strong dependency: The security mechanism assumes that the user has configured git-sync, and the rollback capability is missing in a non-version control environment.
- Personality status black box: The judgment logic of "high-risk personality status" is not detailed in the document, and there is insufficient transparency.
- Gemini API coupling: Exposed in environment variables
GEMINI_API_KEYIt is suggested that there is an external AI call, but the security report does not detect the actual network request. There may be unenabled features or future version dependencies.
Suitable for the crowd
- need7×24 hours autonomous operation and maintenanceUnattended Agent system
- PursueMinimalist dependencies, auditable evolutionsecurity-sensitive scenarios
- Advanced developers who are familiar with Git workflow and can accept the radical concept of "adapt or die"
- Not suitable: A strict compliance environment that requires fully explainable AI decisions, and a traditional operation and maintenance team without version control
General risks
|Risk items|level|illustrate|
|--------|------|------|
|command execution| Low |child_process uses whitelisting and timeout control|
|path traversal| Info |It is recommended that future versions add `isPathSafe` verification|
|Configuration misuse| Medium |Mad Dog Mode is enabled by default, and newbies may deploy it without review.|
|Evolution out of control| Low |The GEP protocol and parent id tree structure provide theoretical traceability, but actual rollback requires manual intervention.|
Key recommendations: Must be enabled in production environment--reviewmode, and configure a standalone git-sync cron job as a safety net.
metaaiautomationdevelopment-engineeringdevopsbackendproductivity
Copyright and takedown notice: AI Islands curates this page from public information. Skills, code, documents and packages remain the property of their original authors or rights holders. This listing is provided for indexing, research and installation convenience. If you believe any listing or download link infringes your rights, contact ai-islands@streamflowintel.com with proof of ownership, relevant URLs and your request. We will review and remove or adjust the content promptly. Review package permissions, dependencies and safety risks before installing.